Research On Face-Enhanced Video Coding Method In NB-IoT Environment

In recent years,community security surveillance video has played an increasingly important role in the "Safe City Construction Project".With the expansion of the deployment range and the improvement of the definition of video,the rapid growth of monitoring data has brought great pressure to transmission and storage.At present,high-definition surveillance video is mainly accessed by broadband,cellular network and Wi-Fi,and the cost remains high,which restricts the deployment of cameras.The rise of smart-Io T technology represented by NB-Io T has created technical conditions for the expansion of surveillance systems.It has the characteristics of wide coverage,multiple connections,low power consumption and low cost,and has good wireless access adaptability,which can greatly reduce the transmission cost of surveillance video and expand the deployment range.However,the upstream bandwidth of NB-Io T is extremely narrow,only 180 kbps,which cannot transmit the high-definition code stream of surveillance video,which limits the application of Io T technology in surveillance systems.According to the characteristics of security surveillance video,it is a technical problem to be solved that the key criminal investigation information can be guaranteed and the ultra-high-power video compression method can be realized to meet the narrowband requirements.Security surveillance video usually pays more attention to key information such as pedestrians and faces.For human faces,high-resolution image quality needs to be maintained to ensure the recognizability of human faces;For pedestrian targets,more attention is paid to the identification characteristics of behaviors,which requires continuity in the time domain,but does not require high resolution.Therefore,for surveillance videos that focus on human faces,this paper proposes a mixed-resolution encoding algorithm,which encodes the face area of key frames with high resolution,and encodes the background with low resolution,which achieves extremely low bit rate compression while ensuring the face quality.Regarding the proposed method,this article mainly does the following work:(1)Propose a mixed-resolution codec framework for surveillance videoAiming at the extremely low bit rate environment of NB-Io T,a mixed-resolution codec framework for surveillance video that focuses on human faces is designed.On the basis of the general video codec framework,a face detection and tracking module is added,a dual stream encoding mode of face and background is introduced,a mixed code stream is designed,a decoding method is improved,and a low-resolution face super-resolution enhancement module is added.All research contents are carried out under this framework.(2)Propose a mixed-resolution video coding methodIn view of the fact that the importance of the face and background is not equal to the occupied bit rate,and the conventional surveillance video coding method cannot take into account the problem of face quality and bandwidth limitations,this paper proposes a key resolution facial resolution fidelity mixed resolution video coding method.Before encoding,firstly propose a fast face extraction algorithm for surveillance video,optimize the single frame face detection speed through statistical training constraint model,and increase the speed of face separation in video by alternating detection and tracking.Then,a spatial domain decomposition method is proposed to realize background sampling and I-frame face sub-frame generation.The background is coded at a low resolution,and the face sub-frame sequence is coded with the corresponding low-definition face as a reference to obtain mixed bit-stream information,which can be decoded to synthesize high-resolution face images of key frames.Experiments show that compared with the mainstream ROI coding method(2016,Liao),the PSNR of the key frame face is increased by an average of 4.8 d B,the subjective quality of the background is better,the coding speed is increased by 4.5 times,and it is more suitable for narrow-band extremely low bit rate environments.(3)Propose face super-resolution algorithm based on key frame referenceAiming at the problem that the existing reference image super-resolution method does not make full use of the face structure prior,and lacks the similarity constraints of face features,resulting in poor recognition of generated images,this paper proposes a dual-stream convolutional neural network based on face feature retention constraints To realize the face over-score process based on high-definition reference.Experiments show that compared with SRTNT(2019,Zhang),the proposed method improves the average PSNR of faces by 1.92 d B and SSIM by an average of 0.02.In the mixed resolution coding framework,it can significantly improve the quality of P frames and improve the decoding of video faces visual fluency.(4)Design video coding system for face monitoring under NB-Io T environmentBased on the above research,this paper further expands the application-level simulation design software.By building a face monitoring simulation system under NB-Io T environment,the effectiveness and practicability of the algorithm in this paper are verified.The interactive operation of the interface module is realized through MFC,the real-time media data stream is transmitted by the RTP protocol,and multiple platforms operate in concert,simulating a narrow-band environment to realize the encoding,decoding,and enhanced recognition processes.The experimental results show that the simulation system has high compression efficiency for 1080 p high-definition surveillance video,and the face recognition rate is close to the original high-definition sequence,which is practical in the NB-Io T environment.In summary,this paper focuses on face-focused surveillance video in the NB-Io T environment,and proposes a mixed-resolution encoding method to achieve fidelity encoding of key frame face resolution,non-key frame face super-resolution enhancement,and low-resolution background degradation The hybrid coding mode achieves high-magnification compression and good face quality,and is unique and practical in the NB-Io T environment.